DetectBERT

DetectBERT: Towards Full App-Level Representation Learning to Detect Android Malware

Environment Setup

To replicate our experiments and use DetectBERT, ensure your environment meets the following requirements:

• Java: 11.0.11
• Python: 3.7.11
• Libraries:
  • numpy: 1.21.6
  • torch: 1.12.1
  • torchvision: 0.2.2
  • torchmetrics: 0.3.2
  • tensorboard: 2.9.1
  • nystrom_attention: 0.0.11
  • scikit-learn: 1.0.2

Data Preparation

Before training the model, class-level DexBERT embeddings for APKs must be generated:

cd data
python GenDexBertEmbeddings.py

Model Training and Evaluation

To train and evaluate DetectBERT:

1. Configure the aggregation method and hyperparameters in model/config.yaml.
2. Execute the training script:

cd model
python main.py